top of page

Blog posts

Lithuanian Llama2 for Assistive Writing

The ABILITY Team



lt-llama2 performance graph (loss vs training steps)
lt-llama2 performance graph - loss vs training steps (image source: https://arxiv.org/pdf/2408.12963)

In late 2024, researchers from neurotechnology have released the preprint of their research on a finetuned Llama2 for Lithuanian language.

In their paper, researchers propose and describe the first open large language models (LLMs) for the Lithuanian language. The authors provide an accompanying question/answer (Q/A) dataset and translations of popular LLM benchmarks, as well as a detailed description of the proposed LLMs and their training process. Additionally, they conduct an empirical evaluation comparing the perplexities of the proposed LLMs with those of other modern open LLMs and benchmark the proposed model against language understanding tasks to determine the importance of high-quality pretraining datasets in achieving efficient model performance on these benchmarks.

Authors have also released the code base for this Lithuanian LLM at Hugging Face.

3 views0 comments

Recent Posts

See All

Demoing ABILITY at DBSV

"We’re excited to share that on January 25th, we were at the Aura-Hotel Saulgrub with the Deutscher Blinden- und Sehbehindertenverband e....

Comentarios

Obtuvo 0 de 5 estrellas.
Aún no hay calificaciones

Agrega una calificación

Human Centred, Multisensory Device Creation

Contact us

If you want to contact the ABILITY Coordination Team, please send an email to:

sabrina.paneels@cea.fr

​

In addition, we encourage you to meet the ABILITY Management Board so you can redirect your question accordingly.

Sign up to get the latest news on our product.

Thanks for subscribing!

This project has received funding from the European Union’s Horizon research and innovation programe under grant agreement Nº 101070396

EU flag
bottom of page